Identifying GDPR Privacy Violations Using an Augmented LSTM: Toward an AI-based Violation Alert Systems

Abstract

After introducing the General Data Protection Regulation (GDPR), it becomes critical to preserve data privacy of individuals and organizations and detect any violations or data collection practices that do not comply with the GDPR articles. However, analyzing privacy incidents, then identifying the consequences and fines require significant effort and time from law enforcement authorities. Additionally, organization need systems that check whether data collections practices, and data processing mechanisms comply with diverse GDPR articles. In this paper, we proposed an approach to identify GDPR violations based on the recent privacy incidents and the semantic similarity of such incidents with the terminology used in different articles. Our approach is driven by both text summarization and deep learning techniques. We used Labeled Topic Modeling approach to identify topics associated with specific types of violations that correspond to different articles. We then used the identified feature to train and test a Long Short-Term Memory(LSTM) Deep Learner that identifies potential violations given textual descriptions. Our approach is compared to conventional text modeling techniques. The result demonstrates a promising accuracy of the proposed approach.